Writerslogic at the CLEF 2026 SimpleText Track: Multi-Candidate LLM Simplification and Stacked Complexity Spotting
arXiv cs.CL · 3 Oct 2026, 02:44 AEST
Writerslogic participated in the CLEF 2026 SimpleText shared task, focusing on text simplification and complexity spotting. Their submissions achieved top rankings in both tasks, utilizing advanced models for effective simplification and error detection.
Key facts
- For Task 1, we develop a multi-candidate generation pipeline using GPT-4o-mini that produces five simplification candidates per sentence at varying temperatures, then selects the best candidate using a reference-free scoring heuristic that rewards compression, source word retention, Cochrane Plain Language Summary vocabulary usage, and lexical simplicity.
- On Task 1.1 (sentence-level simplification), our Claude Sonnet 4 submission achieves SARI 47.43 and BLEU 14.21, the top-ranked sentence-level system (3rd on the combined Task 1 leaderboard, behind two document-level submissions).
- For Task 2, we fine-tune a DeBERTa-v3-large NLI model on 350K labeled (source, sentence) pairs, framing hallucination detection as natural language inference.
- On Task 2.1 (binary overgeneration identification), our fine-tuned DeBERTa system achieves 0.8081 document-level macro F1 (0.8085 in our best ensemble), the top-ranked entry within the identification track and 2nd among teams overall, behind AIIR Lab (0.8197).
- We evaluate both tasks on English and multilingual biomedical text from Cochrane systematic reviews.
Read aloud
Writerslogic participated in the CLEF twenty twenty six SimpleText shared task, focusing on text simplification and complexity spotting. Their submissions achieved top rankings in both tasks, utilizing advanced models for effective simplification and error detection. For Task one, we develop a multi-candidate generation pipeline using G P T four oh mini that produces five simplification candidates per sentence at varying temperatures, then selects the best candidate using a reference-free scoring heuristic that rewards compression, source word retention, Cochrane Plain Language Summary vocabulary usage, and lexical simplicity. On Task one point one sentence-level simplification, our Claude Sonnet four submission achieves SARI forty seven point four three and BLEU fourteen point two one, the top-ranked sentence-level system 3rd on the combined Task one leaderboard, behind two document-level submissions. For Task two, we fine-tune a DeBERTa-v3-large NLI model on 350K labeled source, sentence pairs, framing hallucination detection as natural language inference. On Task two point one binary overgeneration identification, our fine-tuned DeBERTa system achieves zero point eight zero eight one document-level macro F1 zero point eight zero eight five in our best ensemble, the top-ranked entry within the identification track and 2nd among teams overall, behind AIIR Lab zero point eight one nine seven. We evaluate both tasks on English and multilingual biomedical text from Cochrane systematic reviews.
Sources in this cluster
- Writerslogic at the CLEF 2026 SimpleText Track: Multi-Candidate LLM Simplification and Stacked Complexity Spotting
arXiv cs.CL · 3 Oct 2026, 02:44 AEST